Messages in this thread | | | Date | Mon, 10 Sep 2012 15:46:36 +0100 | From | "Jan Beulich" <> | Subject | Re: [PATCH 2/4] x86/xor: improve XMM register spill/fill |
| |
>>> On 10.09.12 at 16:05, "H. Peter Anvin" <hpa@zytor.com> wrote: > On 09/10/2012 05:38 AM, Jan Beulich wrote: >> +/* >> + * By forcing the alignment beyond the default of 16 bytes, we make the >> + * compiler guarantee the alignment. Passing -mincoming-stack-boundary=3 >> + * (which would have been the better global alternative, as the kernel >> + * never guarantees better stack alignment) isn't permitted on x86-64. >> + */ > > The very latest gcc should handle it, and in fact we compile with > -mstack-alignment=3 if gcc accepts it (if it is not yet upstream it will > be soon.) This affects the validity of this patch.
The comment would be stale with that, but the code should still be fine - it would merely over-align the stack in that case (to 32 bytes when 16 would suffice). Or did you spot something else that I'm missing?
Also, I can't spot any use of -mstack-alignment= in today's tip's arch/x86/Makefile* - where's that hidden?
Jan
| |